3D time-lapse seismics, also known as 4D seismics, have great potential for monitoring the migration of CO2 at underground storage sites. This thesis focuses on time-lapse analysis of 3D seismic reflection data acquired at the Ketzin CO2 geological storage site in order to improve understanding of the reservoir and how CO2 migrates within it. Four 3D seismic surveys have been acquired to date at the site, one baseline survey in 2005 prior to injection, two repeat surveys in 2009 and 2012 during the injection period, and one post-injection survey in 2015. To accurately simulate time-lapse seismic signatures in the subsurface, detailed 3D seismic property models for the baseline and repeat surveys were constructed by integrating borehole data and the 3D seismic data. Pseudo-boreholes between and beyond well control were built. A zero-offset convolution seismic modeling approach was used to generate synthetic time-lapse seismograms. This allowed simulations to be performed quickly and limited the introduction of artifacts in the seismic responses. Conventional seismic data have two limitations, uncertainty in detecting the CO2 plume in the reservoir and limited temporal resolution. In order to overcome these limitations, complex spectral decomposition was applied to the 3D time-lapse seismic data. Monochromatic wavelet phase and reflectivity amplitude components were decomposed from the 3D time-lapse seismic data. Wavelet phase anomalies associated with the CO2 plume were observed in the time-lapse data and verified by a series of seismic modeling studies. Tuning frequencies were determined from the balanced amplitude spectra in an attempt to discriminate between pressure effects and CO2 saturation. Quantitative assessment of the reservoir thickness and CO2 mass were performed. Time-lapse analysis on the post-injection survey was carried out and the results showed a consistent tendency with the previous repeat surveys in the CO2 migration, but with a decrease in the size of the amplitude anomaly. No systematic anomalies above the caprock were detected. Analysis of the signal to noise ratio and seismic simulations using the detailed 3D property models were performed to explain the observations. Estimation of the CO2 mass and uncertainties in it were investigated using two different approaches based on different velocity-saturation models.